JP7303365B2 - Sensor calibration based on string of detected values - Google Patents

Description

The present disclosure relates to vehicle sensors and, according to some aspects, to wireless detection and ranging (radar) transceivers for use with vehicles.

Modern vehicles often include one or more radar transceivers arranged to detect objects in the vicinity of the vehicle. Objects can be stationary objects such as guardrails and roadway safety zones, or they can be moving objects such as other vehicles.

A radar transceiver is configured to estimate the distance between the transceiver and a detected object, eg, based on transmission of a frequency modulated continuous wave (FMCW) signal. The velocity of the detected object relative to the radar transceiver can also be determined based on the Doppler shift of the received waveform.

Some radar transceivers are also arranged to estimate azimuth relative to detected objects in addition to range and relative velocity. This bearing estimation can be accomplished by using an array antenna configured to determine the relative phases of waveform components received at different antenna elements within the antenna array. Azimuth estimation, ie, estimating the relative angle to a detected object with respect to, for example, the boresight direction of a radar transceiver, is generally a difficult problem. One aspect that makes this problem particularly difficult is that accurate azimuth estimation requires careful calibration of the radar transceiver.

式中、ｖ_Ｄは、ドップラー測定から得られた相対速度を表し、ｖ_Ｖは、例えば、全地球測位システム（Global Positioning System、ＧＰＳ）受信機又は車輪速度センサから得られたレーダトランシーバ速度を表す。 A known algorithm for calibrating a radar transceiver to improve heading estimation accuracy is to calculate the relative velocity of a detected stationary object compared to the velocity of the radar transceiver, e.g., the velocity of the vehicle in which the radar transceiver is mounted. including a method using the ratio of Azimuth θ can then be estimated or obtained from the following equation.

where v _D represents the relative velocity obtained from Doppler measurements and v _V represents the radar transceiver velocity obtained, for example, from a Global Positioning System (GPS) receiver or wheel speed sensor. .

A problem with the above approach is that the calibration accuracy strongly depends on the vehicle speed accuracy. An error of only a few percent in the determination of _vV can result in an error in the calibration of orientations such as numbers. Such errors are often unacceptable in safety-critical applications such as Level 4 (L4) autonomous driving, advanced driver assistance systems (ADAS), and the like.

A more accurate method for calibrating radar transceivers and other types of vehicle sensors such as Lidar (Light Detection and Ranging) sensors, camera sensors, or ultrasonic sensors is needed in terms of orientation. It is said that Such methods should preferably have limited computational complexity to conserve radar transceiver processing resources, which are often constrained.

SUMMARY OF THE DISCLOSURE It is an object of the present disclosure to provide a method for calibrating the angle measurements of vehicle sensors in a vehicle that mitigates at least some of the drawbacks discussed above. It is also an object of the present disclosure to provide corresponding control units, computer program products and vehicles.

This object is obtained by a method for calibrating the angle measurements of vehicle sensors in a vehicle. The method includes obtaining a sequence of detection values associated with the stationary target and determining a corresponding sequence of ground truth values associated with the stationary target, the ground truth values being measured from the vehicle to the stationary target. It is determined based on the minimum detectable distance and based on the trajectory of the vehicle. The method also includes estimating a difference between the sequence of detected values and the sequence of ground truth values, and calibrating the angle measurement based on the estimated difference.

This method achieves angle calibration independent of vehicle speed, which is an advantage. This method instead uses the minimum detectable distance to a stationary object, data that can be obtained with relatively high accuracy. Thus, a robust and accurate method for calibrating vehicle sensors is provided.

This method can be performed on-line while the vehicle is in operation, which is an advantage as it allows continuous maintenance of vehicle sensor calibration over time.

According to some aspects, acquiring includes detecting stationary objects in a target list associated with vehicle sensors. It is an advantage that the method can be applied to regular target lists obtained from vehicle sensors. No major redesign is required in the vehicle sensor to enable the calibration routine.

According to some aspects, obtaining includes obtaining the detection value when the vehicle is traveling in a straight line. By limiting the method to linear, or at least substantially linear driving scenarios, the more complex motion patterns of the vehicle and the effect of such patterns on the final result can be neglected. The complexity of such methods is further limited, which is an advantage as it saves processing resources of vehicle sensors or some associated processing system.

According to some aspects, obtaining includes obtaining a detection value when the vehicle enters a predetermined calibration area. Some regions may be more suitable than others for performing the proposed methods. Such areas may be identified in advance and this information may be communicated to the vehicle. Therefore, by preferably performing the method on these types of calibration regions, the overall calibration performance can be improved.

According to some aspects, obtaining includes rejecting detection values associated with sidelobes of vehicle sensors. For example, sensor readings, such as radar readings, often include sidelobes that extend on either side of the sensor mainlobe. Sidelobe detection values can degrade the performance of the method. However, by rejecting sidelobe senses, the impact of sidelobe senses can be reduced, which is an advantage.

According to some aspects, the method also includes adjusting the ground truth value based on the estimated trajectory of the vehicle. In this way, additional calibration data can be obtained, for example, since the method is not limited to straight-line driving scenarios. By considering vehicle motion in addition to minimum distance, the method is applicable to a wider range of driving scenarios, which is an advantage.

According to some aspects, the method includes filtering differences by a moving center window. In this way the impact of outlier measurements is reduced, which is an advantage.

According to some aspects, calibrating includes weighting based on the sequence of detected values obtained. Some sequences of detected values are likely to be better than others, ie, likely to yield more accurate calibration data. For example, a data sequence acquired in a particular region known to be suitable for calibration based on the proposed method can be compared with a sequence acquired in a region known to be unsuitable for calibration. can result in higher weights. In this way the robustness of the proposed method is improved.

According to some aspects, calibrating includes weighting based on vehicle trajectory while obtaining detection values. The most likely straight-line driving scenario is associated with higher calibration performance compared to the more complex calibration scenarios where the vehicle steers. Considering the motion of the vehicle improves the performance of the method.

Also disclosed herein are vehicles, control units, and computer program products relating to the above advantages.

The present disclosure will be described in more detail with reference to the accompanying drawings.
1 shows a vehicle with a sensor system; 4 is a graph showing exemplary sensor readings; 5 is a graph showing exemplary detection distances; 4 is a flow chart illustrating a method; 4 shows an exemplary control unit; 4 shows an exemplary control unit; 1 illustrates an exemplary computer program product;

Aspects of the present disclosure are described more fully below with reference to the accompanying drawings. However, different arrangements, devices, systems, computer programs, and methods disclosed herein may be embodied in many different forms and should not be construed as limited to the aspects set forth herein. shouldn't. Like numbers in the drawings refer to like elements throughout.

The terminology used herein is for the purpose of describing aspects of the disclosure only and is not intended to be limiting of the invention. As used herein, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise.

FIG. 1 shows a vehicle 100 with vehicle sensor systems 110,130. The vehicle moves in a direction with velocity vector V _ego . The vehicle sensor system includes vehicle sensors 110 and sensor data processing system 130 . An optional storage module may also be included in the vehicle sensor system. The storage module may store map data that includes information about the surrounding environment of vehicle 100 . Storage modules are discussed in more detail below.

The sensor data processing system is arranged to control the operation of the vehicle sensors 110 and obtain data from the vehicle sensors 110 , such as detection values or data points corresponding to objects 140 near the vehicle 100 .

Vehicle sensor 110 may comprise a single sensor unit or multiple sensor units, potentially of different types and/or different configurations. For example, vehicle sensors 110 may include any of Lidar (light detection and ranging) sensors, radar sensors, camera sensors, or ultrasonic sensors. Vehicle sensors 110 may also include any of TOF camera systems, structured light systems, and stereoscopic systems. Although the focus of this disclosure is on radar-type sensors, the techniques are also applicable to other types of sensors, particularly Lidar sensors.

According to one example, vehicle sensor 110 is arranged to generate a number of data points that include coordinates in a relative coordinate system, ie, relative to the position and orientation of vehicle sensor 110 . The coordinates may be Cartesian or polar. Vehicle sensors 110 may also be arranged to produce raw data, or more refined feature vectors extracted from sensor data.

The data points may also include additional information such as signal strength values or signal-to-noise ratio (SNR) measurements.

One or more timestamps may also be associated with information passing between vehicle sensors 110 and sensor data processing system 130 . Timestamps may be assigned to each data point or may be common to a set of data points.

Vehicle sensor systems 110 , 130 may be connected via wireless link 150 to remote server 160 , which may be included in remote network 170 .

Vehicle sensor 110 is associated with field of view 120 . If the vehicle sensor 110 is a forward radar transceiver, the boresight direction 121 of the sensor 110 is often coincident with the centerline of the field of view 120 . If the vehicle sensor is instead configured as a side sensor, the boresight direction may be oriented at some other angle compared to the vehicle 100 direction.

Some radar transceivers are equipped with antenna arrays that allow estimating not only the range to an object, but also the bearing, ie the measured angle 122 from the sensor boresight direction 121 to the detected object 140 .

As noted above, such angle estimates are often associated with relatively large errors. FIG. 2 shows an example of radar transceiver measurement data 200 . The position of the radar transceiver is indicated by a star 240 with coordinates (0,0). A sequence of detection values 210 is generated for a stationary object in the vicinity of the radar transceiver. Here the vehicle traveled along a straight line, ie along the x-axis in FIG. However, due to calibration errors, there is a spread in the azimuth estimates for the detected objects, appearing as some deviation in the upper portion 215 of graph 200 . It is desirable to compensate for this error by calibrating the radar transceiver 110 .

To perform calibration, it is proposed herein to generate a series of ground truth detections 220 by estimating what the sequence of detections would look like in the absence of angular error. In particular, this estimation or prediction is based on the minimum distance detected between the vehicle and the obstacle. For many types of radar, the minimum distance d _min detected between a vehicle and an obstacle corresponds to the time at which the vehicle passes the obstacle, i.e. the obstacle is located directly beside the vehicle. This point can then be extrapolated based on vehicle motion, ie, how the vehicle moves to that point when the minimum distance is detected. For example, if the vehicle is traveling in a straight line, a reasonable sequence of ground truth values can be estimated as a line of detection values starting from point 221 directly beside the vehicle at the minimum detection distance. The straight line extends parallel to the direction the car was heading up to the time the minimum distance detection was made.

Reference is also made below to FIG. 3, which schematically illustrates exemplary distances in a sequence 310 of detection values associated with several stationary objects. A minimum distance d _min is indicated. According to some aspects, the distance measurements 310 are filtered 320 or smoothed to suppress noise prior to determining the minimum distance. The filtered range estimate is shown in dashed lines in FIG.

The minimum detected distance d _min for a stationary object is assumed to be right beside the vehicle, and this location 221 is the reference point for the sequence 220 of ground truth values. The vehicle path is then backtracked to generate a series of ground truth values. If the vehicle were traveling along a straight line, the ground truth values would also be distributed along a straight line, parallel to the vehicle's trajectory, as shown in FIG. However, if the vehicle follows some other path, this path will be reflected in a series of ground truth values. For example, a vehicle that has performed a turn before a minimum distance is detected will have a series of curvatures corresponding to backtracking of the vehicle motion based on, for example, the vehicle turn rate or yaw rate, and potentially also the vehicle speed. yields a ground truth value of

Having the actual radar detections 210 and the corresponding estimated ground truth values 220, the calibration error can be determined by comparing the detections 210 with the corresponding estimated ground truth values. The difference between these two data sets can be used to correct the radar transceiver data and thus calibrate the radar bearing. Note that the correction value 230 is shown in FIG. 2 and the error is greatly reduced.

According to some aspects, sensor samples are obtained at regular time intervals rather than regular geometric intervals. In this case, the distance between ground truth values may need to vary as a function of vehicle speed up to the point at which the minimum distance is detected.

It is also possible to interpolate between the estimated ground truth values to compensate for time offsets between the sensor samples and the estimated ground truth values.

In summary, referring to FIG. 4, a method for calibrating angle measurements 122 of vehicle sensors 110 in vehicle 100 is disclosed herein. The method includes obtaining a sequence 210 of detection values associated with a stationary target 140 (S1). There are several known methods for classifying sensor readings, such as radar readings, as stationary (as opposed to being moving objects). For example, classification can be based on vehicle motion. A stationary object stays in the same geographical position as the vehicle moves, whereas a dynamic object has its own trajectory of motion. By comparing vehicle motion to sensor detections, stationary objects can be identified over time from at least. A stationary object may also be identified based on data from a tracking algorithm configured to detect the object and track the object over time. Note that the classification problem involves time averaging and is therefore not overly sensitive to errors in vehicle speed, for example.

According to some aspects, acquiring includes detecting stationary objects in a target list associated with vehicle sensor 110 (S11). Using stationary objects avoids the problems associated with estimating the relative motion between two moving objects. Therefore, calibration performance is improved.

According to some aspects, obtaining includes obtaining the detection value when the vehicle 110 is traveling in a straight line (S12). As illustrated and described in connection with FIG. 2, straight-line driving scenarios allow particularly efficient generation of ground truth data.

According to some aspects, obtaining includes obtaining detection values when vehicle 110 enters a predetermined calibration area (S13). Some straight courses of road may be more suitable for performing the method than others. For example, some regions may contain suitable stationary objects that give strong sensor detection values that can be used in the method. Such suitable stationary objects may also be placed by an operator or other third party to facilitate sensor calibration. Referring to FIG. 1, remote server 160 may be configured to store information regarding suitable calibration regions and communicate this information to vehicle 100 over wireless link 150 . The vehicle may include a GPS receiver or the like that allows detecting when the vehicle enters an area where it is recommended to perform the method, and good results can be expected. For example, a suitable calibration area may include a straight section of road with good visibility and suitable stationary targets. It may be preferable to perform the calibration in an environment with a limited number of robust stationary targets to avoid problems associated with target mix-ups and the like.

According to some aspects, obtaining includes rejecting detections associated with sidelobes of vehicle sensors 110 (S14). Sidelobe detections are not obtained based on the boresight mainlobe direction and are therefore associated with larger angular errors.

The method also includes determining a corresponding sequence 220 of ground truth values associated with the stationary target 140 (S2). In particular, the ground truth value is determined based on the minimum detectable distance from vehicle 110 to stationary target 140 and based on the trajectory of vehicle 110 . The minimum detection distance was described above in connection with FIG.

According to some aspects, the method also includes adjusting (S21) the ground truth value based on the estimated trajectory of the vehicle 110 . This method does not require driving in a straight line, as was the case in the example described in connection with FIG. 2, but rather any vehicle trajectory can be used in the method. The ground truth values are then adjusted to account for the vehicle path during the collection of sensor readings up to, and potentially beyond, the point at which the minimum distance is detected. The reference point for the string of sensor readings is again the point with the smallest distance, which can be assumed to be the position directly beside the vehicle (at least for front radar systems).

As mentioned above, the method also includes estimating (S3) a difference 250 between the sequence of detected values 210 and the sequence of ground truth values 220, and calibrating the angle measurement 122 based on the estimated difference 250. (S4).

According to some aspects, estimating the difference includes filtering (S31) the difference with a moving central window. A moving median window replaces individual results with the median over the window. The essence of the median window or filter is to summarize the signal entry by entry and replace each entry with the median of its neighboring entries. The pattern of neighbors is called a "window" and slides entry by entry throughout the signal. For 1D signals, the most obvious windows are the first few previous entries and later entries. In this way measurement errors and outlier results are suppressed. Of course, other types of windows can also be applied in this context.

According to some aspects, calibrating includes weighting (S41) based on the sequence of detected values obtained. Some sequences of detection values may, for example, be associated with high signal-to-noise ratios and thus may be more dependent on weaker detection values.

According to some aspects, calibrating includes weighting (S42) based on vehicle trajectory while obtaining detection values. For example, straight lines often give better results than more complex vehicle steering trajectories.

According to some aspects, calibration includes filtering (S43) with a growing memory filter. This filtering reduces the effects of noise and thus improves the final result. Filtering need not be by extended memory filters, and most known noise suppression filters can be used effectively to improve calibration results.

FIG. 5 schematically shows control unit 130 for calibrating angle measurements 122 of vehicle sensors 110 in vehicle 100 . The control unit
an acquisition unit S1x configured to acquire a sequence 210 of detection values associated with a stationary target 140;
a determining unit S2x configured to determine a sequence 220 of corresponding ground truth values associated with the stationary target 140, the ground truth values being based on the minimum detected distance from the vehicle 100 to the stationary target 140; a determining unit S2x determined based on the trajectory of the vehicle 100;
an estimation unit S3x configured to estimate a difference 250 between the sequence of detection values 210 and the sequence of ground truth values 220;
a calibration unit S4x configured to calibrate the angle measurement 122 based on the estimated difference 250;

FIG. 6 schematically illustrates the components of radio transceiver 130 in terms of some functional units, according to one embodiment described herein. The processing circuitry 610 is capable of executing software instructions stored in a computer program product, e.g. provided using any combination of one or more of processors (digital signal processors, DSPs) and the like. Processing circuitry 610 may further be provided as at least one application specific integrated circuit (ASIC) or field programmable gate array (FPGA). The processing circuit thus comprises a plurality of digital logic components.

In particular, processing circuitry 610 is configured to cause system 130 to perform a series of actions or steps. For example, storage medium 630 may store a set of actions, and processing circuitry 610 may be configured to retrieve the set of actions from storage medium 630 and cause system 130 to perform the set of actions. A set of operations may be provided as a set of executable instructions. Accordingly, processing circuitry 610 is configured to perform the methods disclosed herein.

Storage medium 630 may also comprise persistent storage, which may be, for example, any one or combination of magnetic memory, optical memory, solid-state memory, or remotely mounted memory.

Sensor signal processing system 130 further comprises an interface 620 for communicating with at least one external device. Accordingly, interface 620 may comprise one or more transmitters and receivers with analog and digital components and a suitable number of ports for wireless communication.

Processing circuitry 610 controls system 130 by, for example, sending data and control signals to interface 620 and storage medium 630 , receiving data and reports from interface 620 , and retrieving data and instructions from storage medium 630 . controls the general behavior of Other components of the control node, as well as related functionality, have been omitted so as not to obscure the concepts presented herein.

FIG. 7 depicts a computer program product 700 containing computer-executable instructions 710 for performing any of the methods disclosed herein.

Claims (12)

A method for calibrating angle measurements (122) of vehicle sensors ( 110) in a vehicle, performed by a control unit (130) in the vehicle , the method comprising:
Obtaining (S1) a sequence of detection values (210) associated with a stationary target (140) , comprising:
detecting (S11) a stationary object in a target list associated with said vehicle sensor (110);
acquiring a detection value while the vehicle (100) is running (S12);
including
determining (S2) a sequence of corresponding ground truth values (220) associated with said stationary target (140), said ground truth values determined based on said sequence of detected values (210); determined based on a minimum detectable distance from the vehicle (100) to the stationary target (140) and based on a trajectory of the vehicle (100);
estimating (S3) a difference (250) between the sequence of detection values (210) and the sequence of ground truth values (220);
calibrating (S4) said angle measurement (122) based on said estimated difference (250). The method of claim 1 , wherein said obtaining further comprises obtaining (S13) a detection value when said vehicle (100) enters a predetermined calibration area. 3. The method of claim 1 or 2 , wherein the obtaining further comprises rejecting (S14) sidelobe related detections of the vehicle sensor (110). A method according to any preceding claim, comprising adjusting (S21) the ground truth value based on an estimated trajectory of the vehicle ( 100 ). Method according to any one of the preceding claims, wherein estimating the difference (250) comprises filtering ( S31 ) the difference with a moving median window or a noise suppression filter. A method according to any one of claims 1 to 5 , wherein said calibrating comprises weighting (S41) based on the sequence of said detected values obtained. 7. The method of claim 6 , wherein calibrating includes weighting (S42) based on the trajectory of the vehicle while obtaining the sensed values. A method according to any preceding claim, wherein said calibrating comprises filtering ( S43 ) by an extended memory filter or a noise suppression filter. The method according to any one of the preceding claims, wherein said vehicle sensors 110 comprise lidar sensors, ie light detection and ranging sensors, radar sensors, camera sensors, or ultrasonic sensors. A computer program , for carrying out the method according to any one of claims 1 to 9 , when running on a computer . A control unit (130) for calibrating angle measurements (122) of vehicle sensors (110) in a vehicle (100), said control unit (130) comprising:
an acquisition unit (S1x) configured to acquire a sequence of detection values (210) associated with a stationary target (140) , comprising:
detecting (S11) a stationary object in a target list associated with said vehicle sensor (110);
The acquiring means acquiring the detection value while the vehicle (100) is running (S12).
and
a determining unit (S2x) configured to determine a corresponding sequence of ground truth values (220) associated with said stationary target (140), said ground truth values being said sequence of detected values (210); a determining unit (S2x), determined based on the minimum detectable distance from the vehicle (100) to the stationary target (140) determined based on and based on the trajectory of the vehicle (100);
an estimation unit (S3x) configured to estimate a difference (250) between the sequence of detection values (210) and the sequence of ground truth values (220);
a calibration unit (S4x) configured to calibrate said angle measurement (122) based on said estimated difference (250). A vehicle (100) comprising the control unit (130) according to claim 11 .

Images